Big Tech Q1 capex hits $133B, locking in a structural depreciation drag through 2030. Memory Big 3 prioritize commodity DRAM over HBM as epi wafer supply tightens. Apple signals SoC capacity is the true unit constraint, not memory.
Big Tech spent $133B in Q1 capex, up 70% y/y. Combined depreciation hit $41.6B in the quarter. Server equipment depreciates over 5-6 years, meaning today's record spend locks in an earnings drag through 2030. Consensus estimates peg annual depreciation for MSFT, AMZN, META, and GOOGL exceeding $430B within five years, against $372B of combined net income last year. They already extended useful lives from ~4 years to 5-6, so that accounting lever is largely spent. Alphabet (GOOGL) is nearing Nvidia's $4.85T market cap, currently sitting at $4.65T, driven by cloud and AI demand proving the ROI thesis. It is unclear if capex treadmill commodity token serving will be rewarded by the market forever, as being a 40% capex intensity business hasn't been rewarded over history, though this cycle might differ.
↳ Goldman Sachs advises long hyperscalers, short semis, arguing compressed multiples price in ROI skepticism, while chip upside is already largely rewarded. The best scenario for this trade is hyperscalers demonstrating positive ROI, driving multiple expansion while semis face capex reduction headwinds.
The supply-demand balance for epitaxial wafers is tightening quicker than expected, leading to turning wafer ASPs. AI's impact on the silicon wafer market is no longer a rounding error. Models estimate leading-edge logic (7nm and below) wafer demand inflecting next year, reaching nearly 1M wpm in CY28, around 10% of total 300mm equivalent demand. Major wafer makers GlobalWafers, SUMCO, Shin-Etsu, and Siltronics stand to benefit. However, profitability in commodity memory has become higher than in HBM, giving the Memory Big 3 a stronger incentive to prioritize commodity DRAM/NAND production in the near term. Preliminary April Korea semiconductor export figures show DRAM at $9.25B (+15% MoM, +83% QoQ) and SSDs at $3.84B (+20% MoM, +181% QoQ). High-end device ecosystems are driving memory pricing, but memory remains ultra unaffordable for some mobile players. Looking ahead, Turing Award laureate David Patterson flags High Bandwidth Flash (HBF) as the highly likely next bottleneck after HBM, expecting a surge in demand.
↳ For Apple, memory is purely a cost variable pressuring margins rather than a shipment bottleneck. The true constraint remains leading-edge SoC capacity.
Apple's Q2 FY2026 earnings delivered broad positive demand signals for premium consumer electronics and advanced-node semiconductors, but carried negative implications for Android OEMs and Windows PC vendors. Apple can secure enough memory, but cannot ship enough products due to insufficient SoC supply from tight TSMC leading-edge capacity. TSMC's unmatched capital discipline allows its fabs to break even at operating income with just a 25% utilization rate, down from 35-40% previously, as revenue outpaces capex. Meanwhile, MediaTek presents a substantial earnings upside, with Goldman Sachs setting a TP5000 implying 90% upside. Next-generation AI ASIC projects will feature ASPs multiple times higher than the current generation. MediaTek is engaging directly in compute dies design in addition to I/O dies, making the project GM-accretive. To manage the technical and mass-production risks of adopting EMIB-T packaging, MediaTek hired former TSMC packaging VP Douglas Yu.
↳ TSMC assumed customers would not leave, but they did. TSMC is now rushing to fill the packaging gap between CoWoS and other advanced nodes to retain clients.
Monolithic Power Systems reports AI-related demand is broadening beyond GPUs into CPU servers, optical modules, switches, and storage, driving higher-voltage power conversion needs. AXTI's Q1 signals indium phosphide (InP) substrate demand is accelerating faster than qualified supply, shifting customer procurement toward multi-year allocation security. Aixtron posted soft Q1 revenue of €59m but secured €171m in orders. Management sizes the optoelectronics opportunity at 80-100 G10 tools per year, representing roughly a €300-400m annual TAM. Cohu is strategically pivoting toward AI infrastructure, targeting thermal management for AI accelerators and HBM with its Eclipse handler for extreme power density applications.
↳ Low TSMC SoIC yields have forced the industry to lower near-term 2026 Co-Packaged Optics (CPO) shipment forecasts. Teradyne reported 1Q26 revenue of $1.28B and GM of 60.9%, but guided 2Q26 revenue down 6% QoQ to $1.15-$1.25B, entering a transition period. Furthermore, Teradyne acknowledged Advantest's capabilities in CPO testing, challenging assumptions of overwhelming dominance.
Bare metal cloud capacity is completely sold out. Top AMD providers like Vultr, DO, TW are moving massive numbers of MI355x, with zero availability. DigitalOcean is pricing H100s at $3.39 and MI300x at $1.99. In China, Huawei will capture the largest share of the domestic AI chip market in 2026. Sales are leaping at least 60% to roughly $12B, up from $7.5B in 2025, driven by the 950PR chip and an upcoming 950DT upgrade in Q4. Beijing requires tech firms to support domestic chips, while US regulators restrict Nvidia's H200 clearance. Morgan Stanley predicts the China AI chip market will hit $67B by 2030, with 86% supplied domestically. SMIC has also successfully produced early 3-nanometer test wafers.
↳ Applied Digital signed a 15-year lease for a 300 MW critical IT load campus with a second US investment-grade hyperscaler. The deal adds ~$7.5B in contracted value, bringing total contracted revenue to $23B+. Over 50% of revenue is now backed by IG tenants.
Full analysis, implications, relative trade read-throughs, positioning reads and catalysts — delivered by email at 07:00 HKT daily.